Hosting for open AI models
Running an open model such as Llama, Mistral or Qwen yourself keeps prompts and data on hardware you control. The main constraint is memory: the model has to fit in RAM, or in GPU memory when a GPU is listed.
What AI models needs
- Enough memory
- Many CPU cores
- Private by design
3 plans from our catalog fit, from S$12.77 / month
What AI models needs from a host
Quantised 7 to 8 billion parameter models need roughly 5 to 8 GB of memory; larger models need proportionally more. Unless a plan lists a GPU, it runs models on the CPU, which works well for private assistants, batch jobs and embeddings, but answers more slowly than a GPU.
Enough memory
The model must fit in RAM with room for the operating system and your app.
Many CPU cores
Without a GPU, more cores mean faster token generation.
Private by design
Prompts and documents stay on your server instead of a third-party API.
Plans that fit AI models
Chosen from our live catalog by their specifications: ai / llm server. Prices in SGD, +vat 5% where applicable.
-
Why it fits
- Sized for running language models.
- 6 GB RAM and 4 CPU cores for the model and your app.
-
Why it fits
- Sized for running language models.
- 16 GB RAM and 6 CPU cores for the model and your app.
-
Why it fits
- Sized for running language models.
- 24 GB RAM and 8 CPU cores for the model and your app.
Not sure? Use the plan finderAI / LLM hostingDedicated servers
Set up AI models in 4 steps
An outline of what happens after you order. Our engineers can do any step for you; just open a ticket.
Choose a plan below with enough memory for the model you want.
Install Ollama or llama.cpp on a Linux image.
Download a quantised model and test its speed with your own prompts.
Put an authenticated API or chat interface in front of it before exposing it to the internet.
AI models hosting questions
Already running somewhere else?
Our engineers move your site, databases and email to us for free, and you test everything before switching.